agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling…
Edit a talking-head or presenter video into a soft archival editorial film (bright B&W archival B-roll, torn-paper collages, presenter splits, kinetic type with text behind the presenter, light grain and bloom, word-timed captions, transition sound effects), rendered on the GPU
Claude Code
$ npx -y skills add julianromli/ai-skills --skill faiz-editorial-film-edit --agent claude-code
Codex
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a codex -y
Cursor
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a cursor -y
Gemini CLI
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a gemini-cli -y
GitHub Copilot
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a github-copilot -y
OpenCode
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a opencode -y
Windsurf
$ npx skills add julianromli/ai-skills --skill faiz-editorial-film-edit -a windsurf -y
name: faiz-editorial-film-edit description: Edit a talking-head or presenter video into a soft archival editorial film (bright B&W archival B-roll, torn-paper collages, presenter splits, kinetic type with text behind the presenter, light grain and bloom, word-timed captions, transition sound effects), rendered on the GPU with raea. Use for new edits, revisions and variants of such a video, and when the user wants this look on their footage, even if they don't name the skill. Guides first-time users through setup first.
Turn the user's recording into a soft archival editorial film. The speech, clip order, length and format of the source stay the same. Talk to the user in their language; write generation prompts in English.
Requirements: macOS on Apple Silicon (raea's renderer and the macOS system fonts Helvetica Neue, Didot, Menlo).
`SKILL_DIR` below means this skill's own folder (the base directory shown when the skill loads). Use that real path in every command; do not assume `~/.claude/skills/...`.
The skill has three parts:
raea composites and renders every frame from `timeline.json`, and ffmpeg encodes and checks the delivery file.
Before any work, run the read-only check:
python3 $SKILL_DIR/engine/check_env.py
If it does not say READY, or this is the user's first edit with this skill, read [references/first-time-setup.md](references/first-time-setup.md) and go through it with the user:
B-roll preferences, optional SFX links, and the target platform.
Do not start editing until the check says READY, or the user accepts a named limit (for example, "no AssemblyAI key: use my SRT").
The user does not need to write a prompt. Start every new project with this interview. Ask only what the user's message has not already answered, in the user's language, with the AskUserQuestion tool (at most 4 questions per call, 2–4 options each; the user can always type "Other"). Make the questions specific: probe the footage first (`ffprobe` and a contact sheet) so you can show what you found.
**Round 1: footage, platform, captions** (before anything else) 1. *Footage*: which file? If none was given, offer recent videos you can see (for example in ~/Downloads) or ask for the path. Probe it and confirm what you found: "81 s, 2880×2160, 25 fps, one presenter, no cuts. Correct?" 2. *Platform and formats* (multi-select): the source format (always made first, as the master) / 9:16 (Reels, TikTok, Shorts) / 4:5 (feed) / 5:4. Extra formats are made from the approved master (section 2b). 3. *Captions*: burned-in captions + SRT (default) / SRT only / none. Save the answer as `"captions": "burn" | "srt" | "none"` in timeline.json.
**Then start the project and transcribe** (`init_project.py`, then `transcribe.py`, about 1 minute). Tell the user in one line that you are reading the speech so the next questions can be about their video. If there is no AssemblyAI key, ask for an SRT or skip the message options and ask them as free text.
**Round 2: message, spellings, B-roll, assets** (from the transcript) 1. *Message*: 2–3 short summaries taken from the transcript; the user picks one or types their own. 2. *Spellings*: list the names, products and places the transcript heard, and the low-confidence words. "All correct" / "Fix some" (typed). Confirm each correction before it goes into `captions_config.json`. 3. *B-roll* (multi-select): public-domain archives (free) / generated shots (Higgsfield: say the price per shot, ask for a budget) / only my own images / type only. 4. *Own assets*: none / a folder (typed path). Copy them to `PROJECT/assets/brand/`, use `look: "color"` for logos and brand photos, and record them in CREDITS.md.
**Round 3: sound and masks** 1. *Sound effects*: the bundled library (24 clips, licence not verified) / my own SFX links or files / none. 2. *SFX density* (only if SFX): main elements (default) / every element / key moments only. 3. *SFX level* (only if SFX): subtle (~14 dB under the voice) / punchy (~8 dB under). 4. *Text behind the presenter*: masks made locally (free, private) / on fal (cloud, a few cents, needs a fal key).
Each round is one AskUserQuestion call with at most 4 questions. Leave out questions that are already answered or do not apply (no SFX → no density/level), and do not move a question to a later round.
**Then confirm the brief.** Write it back in 5–8 short lines (footage, message, platform, B-roll, assets, SFX, masks, corrections) and ask "Start with this brief?" as one question with "Yes, start" / "Change something". Save it as `PROJECT/BRIEF.md`. Do not start the edit plan before the user says yes.
Later in the project, ask again only for real decisions (a generation budget, a variant to pick, deleting files).
1. Read [style-guide.md](references/style-guide.md). It covers composition, grain, bloom, B&W, fonts, motion, masks, splits and sound. 2. Read [reference-map.md](references/reference-map.md). Look at the 4 photos and 4 frame sheets. For motion, look at the reference MP4s in the given time windows; a still cannot show motion. 3. For revisions, also read [revision-checklist.md](references/revisio
A curated collection of AI agent skills for coding, frontend, debugging, product thinking, and workflow automation.
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling…
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior.…
Reverse-engineer and clone one or more websites in one shot — extracts assets, CSS, and content section-by-section and…
Use this when building a sidebar that collapses to an icon rail, or any element that toggles between shown and hidden. Labels…
Composition patterns for building flexible, maintainable React components. Avoid boolean prop proliferation by using compound…
Guidelines for building production-ready Convex apps covering function organization, query patterns, validation, TypeScript…
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or…
Ask which skill or flow fits your situation. A router over the skills in this repo.